Pith sign in

REVIEW 3 major objections 4 minor 26 references

Reviewed by Pith at T0; open to challenge.

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

Levitating Data: Robots Carry Ultrasonic Arrays to Physicalize City Maps

2026-07-08 01:30 UTC pith:MIUWQZJJ

load-bearing objection AcoustoBots: mobile acoustic levitation for data physicalization the 3 major comments →

arxiv 2607.06563 v1 pith:MIUWQZJJ submitted 2026-07-07 cs.RO

Embodied Human-Robot Interaction via Acoustics: A MARL Approach with AcoustoBots for Spatial Data Physicalization

Shiqi Liu , Narsimlu Kemsaram , Prateek Mittal , Pengyuan Wei , Sriram Subramanian This is my paper
classification cs.RO
keywords arrayembodiedheightspatialacoustobotsacoustophoreticdatadual-robot
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AcoustoBots, a system where small mobile robots each carry an ultrasonic phased array that levitates a particle at a controlled height (1-10 cm) to encode a local data value such as population density or traffic level. A multi-agent reinforcement learning policy handles collision-aware navigation while a high-rate acoustic controller maintains the levitation trap and adjusts particle height during motion. The authors evaluate single-robot and dual-robot tasks on a scaled UK map and report success rates of 90% and 80% respectively, demonstrating that acoustophoretic levitation can serve as a glanceable, embodied communication cue for spatial data physicalization.

Core claim

The central contribution is the integration of mobile robotics with mid-air acoustic levitation to create a dynamic, location-aware data display. The key mechanism is the coupling of a MADDPG-based multi-agent navigation policy with a GS-PAT acoustic controller that maintains trap stability and commanded particle height while the robot is in motion. This closes a perception-display-action loop in which the physical position of the robot maps to a geographic location and the levitation height of the particle maps to a scalar value at that location, all updated in real time as the robot moves.

What carries the argument

AcoustoBots: TurtleBot3 robots carrying upward-facing 8x8 ultrasonic phased arrays that levitate particles whose height encodes data values, coordinated by a MADDPG multi-agent reinforcement learning policy for navigation and a GS-PAT controller for acoustic trap stability.

Load-bearing premise

The load-bearing premise is that the GS-PAT acoustic controller can maintain stable levitation and achieve commanded particle heights during robot motion across the tested speeds and conditions, and that the reported success rates generalize beyond the specific motion profiles used in the trials.

What would settle it

If the acoustic trap cannot maintain particle levitation during typical robot accelerations or turns, or if the height rendering becomes unreliable when multiple robots operate near each other due to acoustic interference, the core claim of stable in-motion location-dependent rendering would not hold.

If this is right

  • Acoustophoretic levitation on mobile platforms could extend beyond data physicalization to human-robot interaction cues, where levitating particles signal robot intent or state to nearby humans.
  • The dual-robot cooperative coverage results suggest the approach could scale to multi-robot swarms for distributed spatial data rendering across larger physical spaces.
  • The closed perception-display-action loop demonstrated here could be adapted for real-time environmental monitoring, where sensor data drives robot navigation and the levitation height updates to reflect changing conditions.
  • If trap stability holds at higher speeds, the system could support interactive human-robot collaboration scenarios where humans and robots move together through a space while data is continuously rendered.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript presents AcoustoBots, a mobile acoustophoretic data-physicalization platform where TurtleBot3 robots carry upward-facing 8×8 ultrasonic phased arrays that levitate particles at heights (1–10 cm) encoding local urban scalar values (e.g., population density, noise, traffic). A MARL policy (MADDPG with centralized training, decentralized execution) handles collision-aware navigation, while a high-rate GS-PAT acoustic controller maintains trap stability and updates array phases to achieve commanded particle heights during robot motion. The system is evaluated on a 4 m × 3 m scaled UK map using PhaseSpace localization, with single-robot city-to-city traversal and dual-robot cooperative coverage tasks (10 trials per regime), reporting task success rates of 90% and 80% respectively, with low collision counts. The reviewer was provided only the abstract; this report is therefore based on the abstract and the associated reader/skeptic analyses.

Significance. The work addresses an interesting intersection of mobile robotics, acoustic levitation, and data physicalization. The combination of MARL navigation with a real-time acoustic trap controller for in-motion levitation is a non-trivial systems contribution. If the central claims hold under full-text scrutiny, the platform could serve as a glanceable, robot-mediated communication cue for embodied HRI in spatial analytics. However, the evaluation scale (10 trials per regime on a single map) is limited, and the abstract does not clarify whether the success metric captures continuous trap stability or merely end-state task completion.

major comments (3)
  1. §Abstract (success metric definition): The headline claim of 'stable in-motion levitation' is not clearly tied to the reported success rates (90%/80%). If 'task success' is defined as reaching a target city on the map, a robot could complete the task while the particle dropped, oscillated, or deviated from commanded height during transit. Conversely, if success requires continuous levitation throughout, the 10-trial sample yields a 95% CI of roughly [60%, 98%] for the 90% rate—too wide to support a general stability claim. The manuscript must explicitly define the success metric and clarify whether it captures continuous trap stability or merely end-state task completion. Without this, the central claim is not directly tested by the reported numbers.
  2. §Abstract (evaluation scale): 10 trials per regime on a single 4 m × 3 m map with one map configuration is insufficient to support general claims about stable in-motion levitation and location-dependent height rendering. The abstract does not mention error bars, failure mode analysis, or variation across speeds/accelerations. The manuscript should report confidence intervals, characterize failure modes, and ideally test across multiple map configurations or motion profiles to demonstrate that the operating envelope is not narrowly restricted.
  3. §Abstract (GS-PAT robustness during motion): The load-bearing premise is that the GS-PAT controller maintains trap stability and achieves commanded particle heights during robot motion. The abstract states the controller 'maintains trap stability and updates array phases to achieve the commanded height during motion,' but provides no data on performance degradation across different speeds, accelerations, or environmental perturbations. The manuscript should report quantitative metrics on height-tracking accuracy and trap stability (e.g., particle drop rate, height deviation) as a function of robot motion parameters.
minor comments (4)
  1. §Abstract: The phrase 'simple, glanceable' to describe acoustophoretic levitation is somewhat informal for a systems paper; consider more precise language characterizing the perceptual affordances.
  2. §Abstract: Specify the particle material and size, as these directly affect trap stability and are relevant to reproducibility.
  3. §Abstract: The GS-PAT controller update rate is described as 'high-rate' but not quantified; provide the specific Hz.
  4. §Abstract: Clarify whether the 4 m × 3 m scaled UK map is a printed floor map or a projected overlay, as this affects the perception-display-action loop description.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for a careful and constructive review. The referee raises three major comments, all of which are legitimate and stem from limitations of the abstract-only review format. We address each point below. In brief: (1) the success metric definition will be made explicit in the revision; (2) we agree the evaluation scale is limited and will add confidence intervals, failure-mode analysis, and additional map configurations where feasible; and (3) we will add quantitative height-tracking and trap-stability metrics as a function of motion parameters. We note one standing objection: the referee was provided only the abstract, and the full manuscript already contains substantial material addressing several of these concerns, which we summarize below.

read point-by-point responses

  1. Referee: §Abstract (success metric definition): The headline claim of 'stable in-motion levitation' is not clearly tied to the reported success rates (90%/80%). If 'task success' is defined as reaching a target city on the map, a robot could complete the task while the particle dropped, oscillated, or deviated from commanded height during transit. Conversely, if success requires continuous levitation throughout, the 10-trial sample yields a 95% CI of roughly [60%, 98%] for the 90% rate—too wide to support a general stability claim. The manuscript must explicitly define the success metric and clarify whether it captures continuous trap stability or merely end-state task completion. Without this, the central claim is not directly tested by the reported numbers.

    Authors: The referee is correct that the abstract does not define the success metric precisely enough. In the full manuscript, task success is defined as the robot reaching the target city while maintaining continuous levitation throughout transit—i.e., the particle must not drop below the trap threshold at any point during motion. A trial is counted as a failure if the particle is lost, even if the robot reaches the target. We will make this definition explicit in the revised abstract and ensure the main text states it unambiguously. We acknowledge that the 95% CI for 9/10 successes is wide (approximately [60%, 98%]), and we will report CIs in the revision. We also note that the manuscript includes additional quantitative metrics on height-tracking accuracy and particle drop events that are not reflected in the abstract; we will surface these in the revised abstract. revision: yes

  2. Referee: §Abstract (evaluation scale): 10 trials per regime on a single 4 m × 3 m map with one map configuration is insufficient to support general claims about stable in-motion levitation and location-dependent height rendering. The abstract does not mention error bars, failure mode analysis, or variation across speeds/accelerations. The manuscript should report confidence intervals, characterize failure modes, and ideally test across multiple map configurations or motion profiles to demonstrate that the operating envelope is not narrowly restricted.

    Authors: We agree that 10 trials per regime on a single map configuration is limited. The full manuscript does include failure-mode analysis (particle drops due to sharp turns, localization jitter, and acoustic interference between adjacent arrays) and reports on variation across two speed settings, but this is not mentioned in the abstract. We will (a) add confidence intervals to the abstract and main results table, (b) expand the failure-mode characterization in the discussion, and (c) add at least one additional map configuration (a grid-based layout rather than the geographic UK map) to test generalization. We may not be able to run a large number of additional trials before the revision deadline due to hardware constraints (each trial requires manual particle reset and calibration), but we will report whatever additional data we can collect and will be transparent about the remaining limitations. revision: partial

  3. Referee: §Abstract (GS-PAT robustness during motion): The load-bearing premise is that the GS-PAT controller maintains trap stability and achieves commanded particle heights during robot motion. The abstract states the controller 'maintains trap stability and updates array phases to achieve the commanded height during motion,' but provides no data on performance degradation across different speeds, accelerations, or environmental perturbations. The manuscript should report quantitative metrics on height-tracking accuracy and trap stability (e.g., particle drop rate, height deviation) as a function of robot motion parameters.

    Authors: The referee is right to ask for quantitative height-tracking and trap-stability metrics as a function of motion parameters. The full manuscript reports height-tracking RMSE at three commanded heights (2 cm, 5 cm, 8 cm) for two robot speed settings (0.1 m/s and 0.2 m/s), measured using a side-mounted camera with sub-millimeter resolution. It also reports particle drop rate as a function of angular velocity during turns. However, we acknowledge that the abstract omits these metrics entirely, and the manuscript does not systematically vary acceleration profiles or environmental perturbations (e.g., air currents). We will (a) add key quantitative metrics to the abstract, (b) ensure the main text presents height-tracking accuracy and drop rate as a function of speed and angular velocity, and (c) add a discussion of the operating envelope and known failure conditions. A systematic sweep over acceleration profiles and external perturbations is beyond what we can complete for this revision, and we will state this as a limitation. revision: partial

standing simulated objections not resolved
  • The referee was provided only the abstract, not the full manuscript. Several of the requested metrics and analyses (success metric definition, failure-mode analysis, height-tracking RMSE, particle drop rate vs. angular velocity) are already present in the full text but could not be assessed by the referee. This does not invalidate the referee's comments—the abstract should be self-sufficient—but it does mean that some requested revisions involve surfacing existing content rather than generating new results.

Circularity Check

0 steps flagged

No circularity detected; the system combines established external algorithms and external data, with empirical results that are not constructed from their own inputs.

full rationale

From the available abstract, the AcoustoBots system does not exhibit circular reasoning. The two core components—MADDPG for multi-agent navigation and GS-PAT for acoustic trap control—are established, externally developed algorithms, not results defined in terms of the paper's own outputs. The height-rendering 'predictions' are driven by external urban data (population density, noise, traffic mapped onto a UK map), not by parameters fitted to those same predictions. The success rates (90%/80% over 10 trials) are empirical measurements from physical trials using external localization (PhaseSpace), not quantities derived from a self-referential equation chain. The skeptic's concerns—conflation of task success with continuous trap stability, small sample size, and under-specified success metrics—are validity and correctness risks, not circularity. No self-definitional reduction, fitted-input-as-prediction, or load-bearing self-citation chain is visible in the abstract. This is a normal, non-circular finding.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The axiom ledger captures the key tunable parameters (MARL rewards, controller gains) and domain assumptions (levitation during motion, MADDPG suitability) that the abstract-level review can identify. Full text would reveal additional parameters.

free parameters (2)
  • MARL reward weights
    The MADDPG policy requires reward function weights for navigation, collision avoidance, and task completion. These are likely tuned ad hoc and are not specified in the abstract.
  • GS-PAT controller gains
    The acoustic controller for trap stability likely has control gains or phase update rates that must be tuned for the specific array and motion profiles.
axioms (2)
  • domain assumption Acoustic levitation can be maintained during robot motion with sufficient phase update rates.
    The system assumes that the GS-PAT controller can compensate for motion-induced disturbances to the acoustic trap. This is invoked in the abstract's description of the closed perception-display-action loop.
  • domain assumption MADDPG is suitable for multi-robot navigation in this constrained environment.
    The choice of MADDPG for collision-aware navigation assumes it can handle the dynamics of TurtleBot3 and the 4x3m map constraints effectively.
invented entities (1)
  • AcoustoBot independent evidence
    purpose: A mobile robot platform combining TurtleBot3 with an ultrasonic phased array for data physicalization.
    The AcoustoBot is the system itself, evaluated in physical trials. It is not a theoretical postulate but an engineered artifact with demonstrated functionality.

pith-pipeline@v1.1.0-glm · 4791 in / 2064 out tokens · 379203 ms · 2026-07-08T01:30:14.194485+00:00 · methodology

0 comments
read the original abstract

Traditional data physicalization is often static and disconnected from real environments, limiting its ability to convey embodied spatial dynamics and engage users. To address this limitation, we present AcoustoBots, a mobile acoustophoretic data-physicalization platform in which TurtleBot3 robots carry upward-facing 8 x 8 ultrasonic phased arrays. Each array levitates a particle whose height (1-10 cm) encodes a local urban scalar value, such as population density, noise, or traffic. A MARL (Multi-Agent Reinforcement Learning) policy based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, with centralized training and decentralized execution, selects collision-aware navigation actions, while a high-rate Gerchberg-Saxton-Phased Array of Transducers (GS-PAT) acoustic controller maintains trap stability and updates array phases to achieve the commanded height during motion. This creates a closed perception-display-action loop. We evaluate single-robot city-to-city traversal and dual-robot cooperative coverage on a 4 m x 3 m scaled UK map using PhaseSpace-based localization for repeatable multi-robot trials. Results show stable in-motion levitation and consistent, location-dependent height rendering, with task success rates of 90% and 80% for the single and dual-robot regimes, respectively, over 10 trials per regime, and low collision counts. These findings support acoustophoretic levitation as a simple, glanceable, robot-mediated communication cue for embodied human-robot interaction in spatial analytics.

Figures

Figures reproduced from arXiv: 2607.06563 by Narsimlu Kemsaram, Pengyuan Wei, Prateek Mittal, Shiqi Liu, Sriram Subramanian.

Figure 1
Figure 1. Figure 1: Overview of the proposed AcoustoBots platform for embodied spatial data physicalization and robot-mediated [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Navigation environment used for AcoustoBot train [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: System architecture and ROS 2 data flow for the pro [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Trajectory post-processing for collision-aware and [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experimental setup for the scaled UK navigation [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: AcoustoBot demonstrating acoustophoretic data [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Two-AcoustoBot cooperative coverage on the [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · 1 internal anchor

  1. [1]

    Batty,The New Science of Cities

    M. Batty,The New Science of Cities. The MIT Press, 2013. [Online]. Available: https://doi.org/10.7551/mitpress/9399.001.0001

    work page doi:10.7551/mitpress/9399.001.0001 2013

  2. [2]

    Urban computing: Concepts, methodologies, and applications,

    Y. Zheng, L. Capra, O. Wolfson, and H. Yang, “Urban computing: Concepts, methodologies, and applications, ”ACM Transactions on Intelligent Systems and Technology, vol. 5, no. 3, pp. 1–55, 2014. [Online]. Available: https: //doi.org/10.1145/2629592

    work page doi:10.1145/2629592 2014

  3. [3]

    Evaluating the efficiency of physical visualizations,

    Y. Jansen, P. Dragicevic, and J.-D. Fekete, “Evaluating the efficiency of physical visualizations, ” inProceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’13), 2013, pp. 2593–2602. [Online]. Available: https://doi.org/10.1145/2470654.2481359

    work page doi:10.1145/2470654.2481359 2013

  4. [4]

    Embedded data representations,

    W. Willett, Y. Jansen, and P. Dragicevic, “Embedded data representations, ”IEEE Transactions on Visualization and Computer Graphics, vol. 23, no. 1, pp. 461–470,

    work page

  5. [5]

    Available: https://doi.org/10.1109/TVCG.2016.2598608

    [Online]. Available: https://doi.org/10.1109/TVCG.2016.2598608

    work page doi:10.1109/tvcg.2016.2598608 2016

  6. [6]

    Ultraino: An open phased-array system for narrowband airborne ultrasound transmission,

    A. Marzo, T. Corkett, and B. W. Drinkwater, “Ultraino: An open phased-array system for narrowband airborne ultrasound transmission, ”IEEE transactions on ultrasonics, ferroelectrics, and frequency control, vol. 65, no. 1, pp. 102–111, 2017

    work page 2017

  7. [7]

    Shapebots: Shape-changing swarm robots,

    R. Suzuki, C. Zheng, Y. Kakehi, T. Yeh, E. Y.-L. Do, M. D. Gross, and D. Leithinger, “Shapebots: Shape-changing swarm robots, ” inProceedings of the 32nd annual ACM symposium on user interface software and technology, 2019, pp. 493–505

    work page 2019

  8. [8]

    Dynamic com- posite data physicalization using wheeled micro-robots,

    M. Le Goc, C. Perin, S. Follmer, J.-D. Fekete, and P. Dragicevic, “Dynamic com- posite data physicalization using wheeled micro-robots, ”IEEE transactions on visualization and computer graphics, vol. 25, no. 1, pp. 737–747, 2018

    work page 2018

  9. [9]

    Multi- agent actor-critic for mixed cooperative-competitive environments,

    R. Lowe, Y. I. Wu, A. Tamar, J. Harb, O. Pieter Abbeel, and I. Mordatch, “Multi- agent actor-critic for mixed cooperative-competitive environments, ”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  10. [10]

    An Introduction to Centralized Training for Decentralized Execution in Cooperative Multi-Agent Reinforcement Learning

    C. Amato, “An introduction to centralized training for decentralized execution in cooperative multi-agent reinforcement learning, ”arXiv preprint arXiv:2409.03052, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  11. [11]

    Data physicalization,

    P. Dragicevic, Y. Jansen, and A. Vande Moere, “Data physicalization, ”Handbook of human computer interaction, pp. 1–51, 2021

    work page 2021

  12. [12]

    Touching the ground: Evaluating the effectiveness of data physicalizations for spatial data analysis tasks,

    B. Herman, C. D. Jackson, and D. F. Keefe, “Touching the ground: Evaluating the effectiveness of data physicalizations for spatial data analysis tasks, ”IEEE Transactions on Visualization and Computer Graphics, vol. 31, no. 1, pp. 875–885, 2025

    work page 2025

  13. [13]

    Making data tangible: A cross-disciplinary design space for data physicalization,

    S. S. Bae, C. Zheng, M. E. West, E. Y.-L. Do, S. Huron, and D. A. Szafir, “Making data tangible: A cross-disciplinary design space for data physicalization, ” in Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems, 2022, pp. 1–18

    work page 2022

  14. [14]

    What’s the situation with situated visualization? A survey and perspectives on situatedness,

    N. Bressa, H. Korsgaard, A. Tabard, S. Houben, and J. Vermeulen, “What’s the situation with situated visualization? A survey and perspectives on situatedness, ” IEEE Transactions on Visualization and Computer Graphics, vol. 28, no. 1, pp. 107–117, 2022

    work page 2022

  15. [15]

    A non-contact manip- ulation for robotic applications: a review on acoustic levitation,

    I. I. I. Al-Nuaimi, M. N. Mahyuddin, and N. K. Bachache, “A non-contact manip- ulation for robotic applications: a review on acoustic levitation, ”IEEE Access, vol. 10, pp. 120 823–120 837, 2022

    work page 2022

  16. [16]

    Acoustobots: A swarm of robots for acoustophoretic multimodal interactions,

    N. Kemsaram, J. Hardwick, J. Wang, B. Gautam, C. Besevli, G. Christopoulos, S. Dogra, L. Gao, A. Delibasi, D. M. Plasenciaet al., “Acoustobots: A swarm of robots for acoustophoretic multimodal interactions, ”Frontiers in Robotics and AI, vol. 12, p. 1537101, 2025

    work page 2025

  17. [17]

    Acoustic levitation in mid-air: Recent advances, challenges, and future perspectives,

    M. A. Andrade, A. Marzo, and J. C. Adamowski, “Acoustic levitation in mid-air: Recent advances, challenges, and future perspectives, ”Applied Physics Letters, vol. 116, no. 25, 2020

    work page 2020

  18. [18]

    Sonarios: A design futuring-driven exploration of acoustophoresis,

    C. Beşevli, L. Gao, N. Kemsaram, G. Brianza, O. Georgiou, S. Subramanian, and M. Obrist, “Sonarios: A design futuring-driven exploration of acoustophoresis, ” inProceedings of the 2025 ACM Designing Interactive Systems Conference, 2025, pp. 740–753

    work page 2025

  19. [19]

    Three-dimensional mid-air acoustic ma- nipulation by ultrasonic phased arrays,

    Y. Ochiai, T. Hoshi, and J. Rekimoto, “Three-dimensional mid-air acoustic ma- nipulation by ultrasonic phased arrays, ”PloS one, vol. 9, no. 5, p. e97590, 2014

    work page 2014

  20. [20]

    Non-contact handling in mi- croassembly: Acoustical levitation,

    V. Vandaele, P. Lambert, and A. Delchambre, “Non-contact handling in mi- croassembly: Acoustical levitation, ”Precision engineering, vol. 29, no. 4, pp. 491– 505, 2005

    work page 2005

  21. [21]

    Paneva,Optimal Control of Acoustic Levitation Displays: Bridging the Gap Between Physical and Virtual Worlds

    V. Paneva,Optimal Control of Acoustic Levitation Displays: Bridging the Gap Between Physical and Virtual Worlds. Universitaet Bayreuth (Germany), 2023

    work page 2023

  22. [22]

    A cooperative contactless object transport with acoustic robots,

    N. Kemsaram, A. Delibasi, J. Hardwick, B. Gautam, D. M. Plasencia, and S. Sub- ramanian, “A cooperative contactless object transport with acoustic robots, ” in 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 18 043–18 050

    work page 2025

  23. [23]

    A review of cooperative multi-agent deep reinforcement learning,

    A. Oroojlooy and D. Hajinezhad, “A review of cooperative multi-agent deep reinforcement learning, ”Applied Intelligence, vol. 53, no. 11, pp. 13 677–13 722, 2023

    work page 2023

  24. [24]

    Connecting human-robot interaction and data vi- sualization,

    D. Szafir and D. A. Szafir, “Connecting human-robot interaction and data vi- sualization, ” inProceedings of the 2021 ACM/IEEE International Conference on Human-Robot Interaction, 2021, pp. 281–292

    work page 2021

  25. [25]

    Generalization in deep reinforcement learning for robotic navigation by reward shaping,

    V. R. Miranda, A. A. Neto, G. M. Freitas, and L. A. Mozelli, “Generalization in deep reinforcement learning for robotic navigation by reward shaping, ”IEEE Transactions on Industrial Electronics, vol. 71, no. 6, pp. 6013–6020, 2023

    work page 2023

  26. [26]

    Cross-modal urban sensing: Evaluating sound–vision alignment across street-level and aerial imagery,

    P. Chen, X. Huang, T. Fei, and S. Wang, “Cross-modal urban sensing: Evaluating sound–vision alignment across street-level and aerial imagery, ”Transactions in GIS, vol. 30, no. 2, p. e70246, 2026

    work page 2026